A NNALES DE L ’I. H. P., SECTION B
J. L. P EDERSEN
G. P EŠKIR
Computing the expectation of the Azéma- Yor stopping times
Annales de l’I. H. P., section B, tome 34, n
o2 (1998), p. 265-276
<http://www.numdam.org/item?id=AIHPB_1998__34_2_265_0>
© Gauthier-Villars, 1998, tous droits réservés.
L’accès aux archives de la revue « Annales de l’I. H. P., section B » (http://www.elsevier.com/locate/anihpb) implique l’accord avec les condi- tions générales d’utilisation (http://www.numdam.org/conditions). Toute uti- lisation commerciale ou impression systématique est constitutive d’une infraction pénale. Toute copie ou impression de ce fichier doit conte- nir la présente mention de copyright.
Article numérisé dans le cadre du programme Numérisation de documents anciens mathématiques
http://www.numdam.org/
265
Computing the expectation of
the Azéma-Yor stopping times
J. L. PEDERSEN
Institute of Mathematics, University of Aarhus Ny Munkegade, 8000 Aarhus, Denmark.
E-mail: jesperl@mi.aau.dk
G. PE0160KIR
Institute of Mathematics, University of Aarhus Ny Munkegade, 8000 Aarhus, Denmark.
(Department of Mathematics, University of Zagreb, Bijenicka 30, 10000 Zagreb, Croatia)
E-mail: goran@mi.aau.dk
Vol. 34, n° 2, 1998, p. 276 Probabilités et Statistiques
ABSTRACT. - Given the maximum process
(St)
==Xr)
associated with a diffusion
((Xt),P,),
and a continuousfunction g satisfying g ( s )
s, we show how tocompute
theexpectation
of theAzéma- Yor
stopping
timeas a function of x. The method of
proof
is based uponverifying
that theexpectation
solves a differentialequation
with twoboundary
conditions.The third
’missing’
condition is formulated in the form of aminimality principle
which states that theexpectation
is the minimalnon-negative
solution to this
system.
It enables us to express this solution in a closed form. The result isapplied
in the case when(Xt)
is a Bessel process and g is a linear function. ©Elsevier,
ParisKey words and phrases: Diffusion, the Azéma-Yor stopping time, the minimality principle,
scale function, infinitesimal operator, the exit time, normal reflection, Ito-Tanaka’s formula, Wiener process, Bessel process.
AMS 1980 subject classifications. Primary 60G60, 60J60. Secondary 60G44, 60J25.
Annales de l’Institut Henri Poincaré - Probabilités et Statistiques - 0246-0203
34/98/02/@ Elsevier, Paris
RESUME. -
Etant
donnés une diffusion((Xt),P,),
son maximum(St)
=~r).
et une fonctioncontinue g
vérifiantg(s)
5, nousmontrons comment calculer
explicitement 1’ esperance
dutemps
d’ arret d’Azema-Yoren tant que fonction de x. La méthode de demonstration utilise le fait que celle-ci est solution d’ une
equation
différentielle avec deux conditions « au bord ». Une troisième conditionsous-jacente
est formulée en terme d’unprincipe
deminimalité, lequel
énonce que cetteesperance
est la solution minimale nonnegative
dusystème.
Cecipermet d’ expliciter
cette solutioncomme une forme fermee. Nous
appliquons
ce résultat au cas ou(Xt )
estun processus de Bessel est g une fonction linéaire. (c)
Elsevier,
Paris1. FORMULATION OF THE PROBLEM
Let be a
non-negative
canonical diffusion with theinfinitesimal
operator
ongiven by
where
and f-L
are continuous functions on(0, oc)
anda2
is furthermorestrictly positive (see [6]).
Assume there exists a standard Wiener process( Bt )
such that for every x > 0The main purpose of this paper is to
compute
theexpectation
of theAzéma-Yor
stopping
time(see [ 1 ] ).
Moreprecisely,
for any continuousfunction g
onsatisfying
0g( x)
x for x >0,
the Azema-Yorstopping
time is defined as followswhere
(St)
is the maximum process associated with(Xt)
started at s > 0. The main aim of this paper is to present a method for
computing
the functionfor 0 x s. Here the
expectation
is taken withrespect
to theprobability
measure
P~
:= under which the process(Xt)
starts at x and the process(St)
starts at s.The motivation to
compute
theexpectation
of suchstopping
times comesfrom some
optimal stopping problems (see [2-4]
and[7]).
In theseproblems
it is of interest to know the
expected waiting
time for theoptimal stopping strategy
which is of the form T9 for some g. In view of thisapplication
we have assumed that the diffusion
(Xt)
and the function s ~g(s)
arenon-negative,
but it will be clear from our considerations below that the results obtained aregenerally
valid.The method of
proof
relies uponshowing
that theexpectation
of thestopping
time solves a differentialequation
with twoboundary
conditions.The third
’missing’
condition is formulated in the form of aminimality principle
which states that theexpectation
is the minimalnon-negative
solution to this system
(see Fig.
1below).
It enables us topick
up theexpectation
among allpossible
candidates in aunique
way. Theminimality principle
is the mainnovelty
in thisapproach (compare
with[2], [3]
and[4]).
In Section 2 the
minimality principle
isformulated,
and in Section 3 the existence anduniqueness
of the minimal solution isproved.
The maintheorem is
proved
in Section4,
and in Section 5 anapplication
of thetheorem is
given.
2. THE MINIMALITY PRINCIPLE
In the first
part
of this section we shall observe that the function~ t-~ solves a differential
equation
with twoboundary
conditions.In the
remaining part
of the section we willpresent
theminimality principle
as the
’missing’ condition,
which will enable us to select theexpectation
of the
stopping
time in aunique
way.In the
sequel
we need thefollowing
definitions and results. The scale function is for x > 0given by
Vol. 34, n° 2-1998.
where
We define as usual the first exit time from an interval
by
for 0 a
b,
and thefollowing
formulas for 0 a x bare well-knownLet g
be a continuous functionsatisfying
0g(x)
x for x > 0 such that = is finite for all 0 ~ s. We will now state the first result. Whenever s > 0 isgiven
andfixed,
the function x -m( x, s )
solves the differential
equation
with the
following
twoboundary
conditionsA first
step
in the direction ofverifying
that(s, s)
~ satisfiesthe system above is contained in the
following
result.LEMMA 2.1. - The
function
s ~m(s, s)
- isC 1
andsatisfies
the
equation
Proof. -
Theproof
isessentially
contained in Lemma 1 and Lemma 2 in[3]. Alternatively,
to obtain a betterfeeling why (2.6) holds,
as wellto derive it in another way, one could use
(2.7)
below with(2.1 )+(2.2)
above to
verify
thatequals
theright-hand
side in(2.6),
thus
showing
that(2.5)
isequivalent
to(2.6),
and then follow the secondpart
of theproof
of Theorem 4.1 below..Let
g( s) x s be given
and fixed. It isimmediately
seen thatand
by applying strong
Markovproperty
we getFrom
(2.1)
and(2.2)
we see thatG(~) -
=s)
andH(x)
=E~
solve thefollowing
well-knownsystems respectively
Consequently, by (2.6)-(2.9)
weeasily verify
that x -m(x, s)
solves thesystem (2.3)-(2.5).
Note
since Tg
may be viewed as the exit timeby
diffusion froman open set, the
equation (2.3)
is well-known and the condition(2.4)
isevident. The condition
(2.5)
is less evident but is known to be satisfied ina similar context
(see [6]
p.118-119).
Unfortunately (x, s)
~--~m(x, s)
is notuniquely
determinedby (2.3)
and the two
boundary
conditions(2.4)
and(2.5).
Thus we need another condition to determine(x, s)
~m( x, s) uniquely.
We formulate the third’missing’
condition in the form of aminimality principle (see Fig.
1below):
The
expectation m(x, s )
is the minimalnon-negative
solution to thesystem (2.3)-(2.5).
3. EXISTENCE AND
UNIQUENESS
OF THE MINIMAL SOLUTIONSince
s )
= 0 for 0 xg ( s )
weonly
need to consider m on9 ( s) x ~
s.Throughout
we shall consider thesystem
Vol. 34, n° 2-1998.
Motivated
by
theminimality principle,
in this section we shall prove the existence(and uniqueness)
of a minimalnon-negative
solution to(3.1 ).
Letus introduce the
following
notation:where D =
~(~~, 6’) ~(~) ~ ~’
6’, s >0~
and D° is the interior of D. The function mbelongs
toC2u (D°) ni(x, s)
is(7~
and s -s)
is
C’
on D° .The main result of this section may be now formulated as follows. If Ji4 is
non-empty
then it contains a minimalelement,
i.e.where the infimum is taken
pointwise.
Combined with the results in Section 2 this will be deduced in theproof
of Theorem 4.1 I belowby using
It6calculus. The
proof
we present here is based upon theuniqueness
theoremfor the first and second-order differential
equations.
For this note that the
uniqueness
theoremimplies
that if m1 and m2belong
to .I~ then either rnl > rn2 or mi rn2 on D°. Let(xo, so)
E Dbe
given
and be a sequence of functions from M such thatso) 1 so) .
Due to the remarkjust mentioned,
the sequence isdecreasing,
and therefore the limit existseverywhere,
i.e.for all
(~B s)
E D. If we can show thatthen
using
theuniqueness
theorem it follows that m =In order to prove
(3.3)
we first show that x ~ (x,8)
solves the differentialequation
in(3.1). By
the instantaneousstopping condition,
and theuniqueness theorem,
can be written aswhere s ~ is a
~‘1-function.
Since is adecreasing
sequence offunctions,
the sequence is alsodecreasing,
and therefore itconverges
pointwise
to a functionA,
i. e.for - Hence ~~ ~
.5)
solves the differentialequation
in(3.1).
Annales de l’Institut Henri Poincaré - Probabilités
Obviously ~
- satisfies the firstboundary
condition(instantaneous stopping)
since each j- t2014~
s)
satisfies this condition.Finally,
toverify
the secondboundary
condition(normal reflection),
notethat
straightforward computations
based on the normal reflection condition and(2.1 )+(2.2)
show that s -An (s)
solves thefollowing
differentialequation
or
equivalently
Applying
the monotone convergencetheorem,
we find that s -A(s)
alsosolves the differential
equation (3.4),
and we can conclude that m E4. THE EXPECTATION OF THE
AZÉMA-YOR
STOPPING TIMES The main result of the paper is contained in thefollowing
theorem.THEOREM 4.1. - Let
((Xt) , Pr)
be thenon-negative diffusion defined
in( 1.1)
and let(St)
be the maximum process associated with(Xt ) defined
in(1.2).
Let g be acontinuous function
on~0, oc) satisfying
0g(x)
~for
x >
0,
and let usdefine
thestopping
timeIf
Mfrom (3.2)
is non-empty, then isfinite
and isgiven by
Vol. 34, n° 2-1998.
where
(x, s)
1---+s)
is the minimal element in Jlit . The converse is also true, and we have thefollowing explicit formula
and a criterionfor verifying
that M is non-emptyfor g(s)
~ s with = 0for
0 ~g(s),
which is validin the usual sense
(if
theright-hand
side in(4.1 )
isfinite,
then so is theleft-hand side,
and viceversa).
Proof. -
For(xo, so)
E Dgiven
andfixed,
consider the setChoose bounded open sets
Gi
CG2
C ... such thatDefine the exit time of the two-dimensional diffusion
(Xt, St)
fromGn by
Note that
Exo,so
00 and anr
T~ as n ~ oo.Let rn be any function in M. Note that m E
C2 ~’-
and(St )
is of bounded variation so that Ito’s formula can beapplied (see
Remark 1 in[5]
p.139).
In this way we
get
Due to the normal reflection condition the last
integral
isidentically
zero. Since the set of those u > 0 for which =
Su
is ofLebesgue
measure zero, and = -1 for
g(s)
x s, we can concludethat
Let be a localization for the local
martingale
Then
by
Fatou’s lemma and theoptional sampling
theorem we getEx0,s0[m(X03C3n,S03C3n)] ~ lim inf Ex0,s0[m(X03C3n039BTk,S03C3n039BTk)] = m(x0,s0)
Thus we have the
inequality
for all n >
1,
andby
monotone convergence it followsFrom this we see that is finite.
By
the results in Section 2 weknow that the function
(x, s)
-1
satisfies thesystem (3.1 ),
andsince m is
arbitrary,
hence we obtainThis
completes
the firstpart
of theproof.
To derive
(4.1 )
note that(2.6)
is a first-order linear differentialequation
whose
general
solution iseasily
found to begiven by
the formulaVol. 34, n° 2-1998.
whenever the last
integral
isfinite,
where C is a real constant.Letting
s - oo we find that C = 0
corresponds
to the minimalnon-negative
solution.
Combining
this with(2.7)
and(2.1 )+(2.2),
we obtain theexplicit
formula
(4.1 ).
Theproof
of the theorem iscomplete..
5. AN EXAMPLE
Let
( (Xt ) , P~ )
denote the Bessel process of dimension 0152, where forsimplicity
we assume that 0152 > 1 but 0152#
2.(The
other cases of 0152could be treated
similarly.)
Thus(Xt)
is anon-negative
diffusion with the infinitesimal operator on(0, oo) given by
(For
more information about Bessel processes see[5].) Let g
be a linearfunction
given by
where 0 A 1. Denote the
stopping time g by
It is our aim in this section to
present
a closed formula for theexpectation
of the
stopping
time Moreprecisely,
denote the functionfor 0 x s. Then our main task is to
compute explicitly
the functionInstead of
using (4.1 ) directly,
we shall rather make use of theminimality principle
within thesystem (3.1 ).
According
to Theorem 4.1 we shall consider thesystem
Annales de l’Institut Poincaré - Probabilités
Let As x s be
given
and fixed. Thegeneral
solution to(5.2)
isgiven by
where s -
A ( s )
and s -B ( s )
are unknown functions.By (5.3)
and(5.4)
we findwhenever
(2 1a)~/(°~~)
A1 ,
whereand C is an unknown constant.
In order to determine the constant C we shall use the
minimality principle.
It is
easily
verified that the minimalnon-negative
solutioncorresponds
toC == O. Thus
by (5.5)-(5.7)
with C = 0 we have thefollowing
candidatefor
when As x s. Hence
by applying
Theorem 4.1 we obtain thefollowing
result. Observe that this
example
is also studied in[2]
and[3].
PROPOSITION 5.1. - Let
((Xt), Px)
be a Bessel processof
dimension astarted at x > 0 under
P~ ,
where a > 1 but c~~
2. Thenfor
thestopping
time Ta
defined
in(5.1 )
we haveVol. 34, n° 2-1998.
Fig. 1. - A computer drawing of solutions of the differential equation (2.6) in the case when
a = 4 and A =
4/5.
The bold line is the minimal non-negative solution (which never hits zero). By the minimality principle proved above, this solution equals ma (8,s)
=]
for all s > 0.
REFERENCES
[1] J. AZÉMA and M. YOR, Une solution simple au problème de Skorokhod. Sém. Prob. XIII.
Lecture Notes in Math. 784, Springer-Verlag Berlin Heidelberg, 1979, pp. 90-115 and 625-633.
[2] L. E. DUBINS L. A. SHEPP and A. N. SHIRYAEV, Optimal stopping rules and maximal
inequalities for Bessel processes. Theory Probab. Appl., Vol. 38, 1993, pp. 226-261.
[3] S. E. GRAVERSEN and G. PE0160KIR, Optimal stopping and maximal inequalities for linear diffusions. Research Report No. 335, Dept. Theoret. Stat. Aarhus (18 pp), 1995. To
appear in J. Theoret. Probab.
[4] S. E. GRAVERSEN and G. PE0160KIR, On Doob’s maximal inequality for Brownian motion.
Research Report No. 337, Dept. Theoret. Stat. Aarhus (13 pp), 1995. Stochastic Process.
Appl., Vol. 69, 1997, pp. 111-125.
[5] D. REVUZ and M. YOR, Continuous Martingales and Brownian Motion. Springer-Verlag.
[6] L. C. G. ROGERS and D. WILLIAMS, Diffusions, Markov Processes, and Martingales; Volume
2: Itô’s Calculus. John Wiley & Sons, 1987.
[7] L. A. SHEPP and A. N. SHIRYAEV, The Russian Option: Reduced Regret. Ann. Appl. Probab.
3, 1993, pp. 631-640.
(Manuscript received April 2I, 1997;
Revised 8, 1997.)
Annales de l’Institut Henri Poirzccire - Probabilités